To test the safety and security of AI, hackers have to trick large language models into breaking their own rules. It requires ingenuity and manipulation – and can come at a deep emotional cost

A few months ago, Valen Tagliabue sat in his hotel room watching his chatbot, and felt euphoric. He had just manipulated it so skilfully, so subtly, that it began ignoring its own safety rules. It told him how to sequence new, potentially lethal pathogens and how to make them resistant to known drugs.

Tagliabue had spent much of the previous two years testing and prodding large language models such as Claude and ChatGPT, always with the aim of making them say things they shouldn’t. But this was one of his most advanced “hacks” yet: a sophisticated plan of manipulation, which involved him being cruel, vindictive, sycophantic, even abusive. “I fell into this dark flow where I knew exactly what to say, and what the model would say back, and I watched it pour out everything,” he says. Thanks to him, the creators of the chatbot could now fix the flaw he had found, hopefully making it a little safer for everyone.

The then Newcastle manager’s on-air blast at Sir Alex Ferguson remains a classic Premier League moment

Premier League history is littered with red letter days and Monday 29 April 1996 will for ever rank among the most memorable. Thirty years on, recollections of the aftermath of Newcastle’s 1-0 victory at Leeds remain vivid. Keith Gillespie’s goal saw Kevin Keegan’s team move three points behind the leaders, Manchester United, with two fixtures remaining.

Before Newcastle’s visit to Elland Road, Sir Alex Ferguson craftily suggested that Leeds and Nottingham Forest – the team Keegan’s players would visit three days later – would not try as hard as they had against his own side. Ferguson also pointedly reminded everyone Newcastle had agreed to provide the opposition for Stuart Pearce’s testimonial by the Trent later in the year. This backdrop dictated that Keegan used a live post-match television interview with Richard Keys and Andy Gray of Sky Sports to claim the moral high ground while also walking straight into Ferguson’s psychological trap.

Everything was in place for Andrea’s dream celebration in Barbados. Then her close friend was denied the right to walk across a British airport to board a connecting flight

Weddings can be complicated to organise, especially when the venue is more than 4,000 miles from home. But Andrea, a Londoner, was confident she and her partner, Josh, had thought of everything when they planned their dream wedding in Barbados for the beginning of May.

The British couple – Andrea of Nigerian and Josh of Bajan heritage – booked a stunning venue, with tropical gardens and spectacular views.

This ‘nonfiction thriller’ takes us through exactly what would happen if Nigel Farage won his dreamed-of majority

For some years now, mainstream British politics has revolved increasingly obsessively around the question of how to stop Nigel Farage. What started a decade ago with Brexit may yet end in a general election that boils down to one question: do you or don’t you want to risk putting this man in Downing Street? That said, we still know surprisingly little about what a Reform government might mean in practice.

Of course, it might never happen. But if it did, what exactly would Farage do with a majority that enabled him to fulfil his wildest dreams? And how well would an unwritten British constitution, still heavily reliant on good chaps voluntarily being good chaps, cope with full-fat populism?

Pace of sea-level rise has turned Outer Banks coastal area into a ‘canary in the coalmine’ for other east coast communities

Moving house has a more literal meaning on Hatteras Island, the slender hook of land that juts off the coast of North Carolina. After a slew of houses toppled spectacularly into the Atlantic Ocean recently, entire buildings are now being lifted on to wheels to flee the rapidly eroding coastline.

Since September, 19 homes have been lost to waves that tore them from their pilings, sending them crashing into other structures like bumper cars before breaking up in the ocean. Spooked homeowners have turned to the unusual services of Barry Crum, a lifelong Hatteras resident who has become the island’s main house mover.

Trump is volatile, capricious and unreasonable – but he belongs to the old world of analogue power. What comes next will be harder to manage

Donald Trump is not impressed by soft power. He respects hard men with military muscle. But he can be moved by pageantry, which is the purpose of King Charles’s visit to Washington this week. Trump is flattered to rub shoulders with majesty. The good vibes are then supposed to radiate warmth through a political relationship that has been chilled by the war in Iran.

It might work, but not for long. Trump’s irritation with Keir Starmer and other European leaders for what he calls cowardice in the Middle East is aggravated daily by evidence that the war is a strategic calamity.

Rafael Behr is a Guardian columnist

Guardian Newsroom: Can Labour come back from the brink?
On Thursday 30 April, join Gaby Hinsliff, Zoe Williams, Polly Toynbee and Rafael Behr as they discuss how much of a threat Labour faces from the Green party and Reform UK – and whether Keir Starmer can survive as leader. Book tickets here or at guardian.live

Reform leader changed his mind about standing as MP after gift from Thai-based crypto tycoon Christopher Harborne

• UK politics live – latest updates

Nigel Farage was given £5m by the crypto billionaire Christopher Harborne shortly before announcing he would stand in the 2024 British general election, the Guardian can reveal.

Farage had stated he did not intend to stand as a prospective MP but U-turned in June 2024, within weeks of receiving the personal gift from the Thailand-based businessman.

A 45 year-old man arrested after two people attacked in north London

Specialist officers from Counter Terrorism Policing are leading the investigation and working with police to establish the full circumstances and any links to terrorism, the Met said in a statement.

Head of counter terrorism policing Laurence Taylor said:

Whilst I must stress this investigation is at an early stage, we are working quickly to understand exactly what happened.

Thank you to those who were in the area at the time and supported the response to this terrible incident.

Our thoughts are with the victims of this horrific attack. We are grateful to officers who swiftly Tasered and arrested the suspect before he could cause further harm.

We are aware of the significant distress and concern this incident is likely to cause in the face of a number of incidents in the local area. A suspect is in custody, and investigators are considering all possible motives.

An investigation is under way and a man has been arrested following a stabbing incident in Barnet.

At 11:16hrs on Wednesday 29 April, officers responded following reports of people stabbed in Highfield Avenue.

Ruling is blow to Office for Students after it issued record fine for allegations over professor’s trans rights views

Sussex University has overturned a £585,000 fine from England’s higher education watchdog after the high court rejected claims that the university breached free speech regulations in a case involving its former professor Kathleen Stock.

The ruling is a damaging blow to the credibility and management of the Office for Students as the court rejected the regulator’s lengthy investigation involving Stock’s resignation in 2021, which followed protests over her views on transgender rights and gender identity.

About 500 officers deployed as part of operation at headquarters of Ahmadi Religion of Peace and Light

Police have raided the headquarters of the Ahmadi Religion of Peace and Light following an investigation into allegations of serious sexual offences, modern slavery and forced marriage.

About 500 officers drawn from across the north-west of England were on the ground on Wednesday morning and made a number of arrests. The officers had warrants to enter the group’s headquarters, which is based out of a former orphanage, as well as other properties in Crewe, Cheshire.